Skip to main content
Glama
deenesjakoruzh
kagisearch

Kagi MCP Server

Official
by kagisearch
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Hybrid

Kagi MCP 服务器

设置说明

在开始之前，除非你只是使用非搜索工具，否则请确保你有权访问搜索 API。它目前处于封闭测试阶段，仅限受邀使用。请联系 support@kagi.com 获取邀请。

首先安装 uv。

MacOS/Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

通过 Smithery 安装

或者，你可以通过 Smithery 为 Claude Desktop 安装 Kagi：

npx -y @smithery/cli install kagimcp --client claude

与 OpenAI 配合使用

Codex CLI

要将 Kagi MCP 服务器添加到 codex cli，你需要使用以下命令：

codex mcp add kagi --env KAGI_API_KEY=<YOUR_API_KEY_HERE> -- uvx kagimcp

这将把配置写入 ~/.codex/config.toml，因此如果你需要更新/轮换你的 API 密钥，请在再次运行 codex 之前在那里更新你的密钥。

Codex CLI 自带内置搜索（通过 --search 标志），但默认情况下是禁用的。因此，为了避免搜索和 Kagi 之间的冲突，只需不要启用它即可。

与 Claude 配合使用

Claude Desktop

// claude_desktop_config.json
// Can find location through:
// Hamburger Menu -> File -> Settings -> Developer -> Edit Config
{
  "mcpServers": {
    "kagi": {
      "command": "uvx",
      "args": ["kagimcp"],
      "env": {
        "KAGI_API_KEY": "YOUR_API_KEY_HERE",
        "KAGI_SUMMARIZER_ENGINE": "YOUR_ENGINE_CHOICE_HERE" // Defaults to "cecil" engine if env var not present
      }
    }
  }
}

Claude Code

使用以下命令添加 Kagi MCP 服务器（设置摘要引擎是可选的）：

claude mcp add kagi -e KAGI_API_KEY="YOUR_API_KEY_HERE" KAGI_SUMMARIZER_ENGINE="YOUR_ENGINE_CHOICE_HERE" -- uvx kagimcp

现在 claude code 可以使用 Kagi MCP 服务器了。但是，claude code 默认自带网络搜索功能，这可能会与 Kagi 冲突。你可以在 claude code 设置文件 (~/.claude/settings.json) 中通过以下设置禁用 claude 的网络搜索功能：

{
  "permissions": {
    "deny": [
      "WebSearch"
    ]
  }
}

提出需要使用工具的查询

例如：搜索查询 "Who was time's 2024 person of the year?"，或摘要查询 "summarize this video: https://www.youtube.com/watch?v=jNQXAC9IVRw"。

调试

运行：

npx @modelcontextprotocol/inspector uvx kagimcp

Related MCP server: sysauto Ask MCP Server

本地/开发设置说明

克隆仓库

git clone https://github.com/kagisearch/kagimcp.git

安装依赖

首先安装 uv。

MacOS/Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows:

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

然后安装 MCP 服务器依赖项：

cd kagimcp

# Create virtual environment and activate it
uv venv

source .venv/bin/activate # MacOS/Linux
# OR
.venv/Scripts/activate # Windows

# Install dependencies
uv sync

与 Claude Desktop 配合使用

使用 MCP CLI SDK

# `pip install mcp[cli]` if you haven't
mcp install /ABSOLUTE/PATH/TO/PARENT/FOLDER/kagimcp/src/kagimcp/server.py -v "KAGI_API_KEY=API_KEY_HERE"

手动设置

# claude_desktop_config.json
# Can find location through:
# Hamburger Menu -> File -> Settings -> Developer -> Edit Config
{
  "mcpServers": {
    "kagi": {
      "command": "uv",
      "args": [
        "--directory",
        "/ABSOLUTE/PATH/TO/PARENT/FOLDER/kagimcp",
        "run",
        "kagimcp"
      ],
      "env": {
        "KAGI_API_KEY": "YOUR_API_KEY_HERE",
        "KAGI_SUMMARIZER_ENGINE": "YOUR_ENGINE_CHOICE_HERE" // Defaults to "cecil" engine if env var not present
      }
    }
  }
}

提出需要使用工具的查询

例如：搜索查询 "Who was time's 2024 person of the year?"，或摘要查询 "summarize this video: https://www.youtube.com/watch?v=jNQXAC9IVRw"。

调试

运行：

# If mcp cli installed (`pip install mcp[cli]`)
mcp dev /ABSOLUTE/PATH/TO/PARENT/FOLDER/kagimcp/src/kagimcp/server.py

# If not
npx @modelcontextprotocol/inspector \
      uv \
      --directory /ABSOLUTE/PATH/TO/PARENT/FOLDER/kagimcp \
      run \
      kagimcp

然后访问 http://localhost:5173 上的 MCP Inspector。你可能需要在 inspector 的环境变量中添加你的 Kagi API 密钥，键名为 KAGI_API_KEY。

高级配置

  • 日志级别可以通过 FASTMCP_LOG_LEVEL 环境变量进行调整（例如 FASTMCP_LOG_LEVEL="ERROR"）

  • 摘要引擎可以使用 KAGI_SUMMARIZER_ENGINE 环境变量进行自定义（例如 KAGI_SUMMARIZER_ENGINE="daphne"）

    • 在此处了解不同的摘要引擎

  • 可能有更安全的方式接入 MCP。一位用户在此处记录了一些详细信息

  • --http CLI 选项可用于开启可流式传输的 HTTP 传输。可以与 --port 和 --host 参数一起使用。

Available Tools

2 tools
kagi_extractA

Extract the content of a web page as markdown using the Kagi Extract API. Use this to read the full content of a page when needed.

ParametersJSON Schema
NameRequiredDescriptionDefault
urlYesThe HTTPS URL of the page to extract content from.

Output Schema

ParametersJSON Schema
NameRequiredDescription
resultYes

TDQS

A4/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

No annotations are provided, so description carries full burden. Only states core behavior without additional context like rate limits, authorization, or side effects. Adequate for a simple read operation.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Two sentences, no fluff, front-loaded with purpose. Every sentence earns its place.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Low complexity with one parameter and existing output schema. Description covers what it does and when to use it, but could mention potential errors or prerequisites for full completeness.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100% and describes the only parameter (url). Description adds no extra meaning beyond the schema, so baseline 3 applies.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

Clearly states the action (extract content), resource (web page), output format (markdown), and API. Distinguishes from sibling tool by focusing on content extraction.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Explicitly says 'Use this to read the full content of a page when needed,' providing clear context. Does not explicitly exclude alternatives, but sibling name implies separation of concerns.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

kagi_search_fetchA

Fetch web results for a query using the Kagi Search API. Use for general search and when the user explicitly tells you to 'fetch' results/information. Results are numbered so that a user may refer to a result by a specific number.

ParametersJSON Schema
NameRequiredDescriptionDefault
afterNoOnly include results published/updated on or after this date (ISO format, e.g., '2024-01-15').
limitNoMaximum number of results per category. In the mixed 'search' workflow this caps each category independently, so the total can exceed this number; in single-category workflows it caps total results.
queryYesA concise, keyword-focused search query. Include essential context for standalone use.
beforeNoOnly include results published/updated on or before this date (ISO format, e.g., '2024-12-31').
lens_idNoApply a Kagi lens to narrow the search to a curated set of sources. Built-in lens IDs: '2' (Academic — education/.edu domains), '1' (Forums — discussion forums across the web), '15' (Programming — official programming language sites and forums), '29' (News 360 — multi-perspective coverage of global news), '120' (Recipes — high-quality recipe sites, English), '107' (Small Web — noncommercial domains and topics). You may also pass a custom lens ID or full URL from https://kagi.com/settings/lenses (only shareable lenses work). Mutually exclusive with 'include_domains', 'exclude_domains', 'time_relative', and 'file_type'; use those args or 'lens_id', not both.
workflowNoType of results to return. Use 'news' for current events and recent reporting, 'videos' for video content (e.g. tutorials, talks), 'podcasts' for audio shows, 'images' for image results, or the default 'search' for general web results. Note that 'search' may return a mix of categories (web, news, videos, images) in one response, like a typical SERP; the other workflows return only their single category.search
file_typeNoRestrict to results with this file type (e.g., 'pdf', 'docx', 'xlsx'). Specify the extension without a leading dot.
extract_countNoNumber of top results to fetch full page content for, inline as markdown.
time_relativeNoRestrict to results published/updated within the last day, week, or month, evaluated server-side. Mutually exclusive with 'after'/'before'.
exclude_domainsNoExclude results from these domains (e.g., ['pinterest.com', 'quora.com']). Overrides any 'site:' operators in the query.
include_domainsNoRestrict results to these domains (e.g., ['docs.python.org', 'github.com']). Overrides any 'site:' operators in the query.

Output Schema

ParametersJSON Schema
NameRequiredDescription
resultYes

TDQS

A4/5.0
Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

The description adds one behavioral trait: 'Results are numbered so that a user may refer to a result by a specific number.' With no annotations provided, the description carries the full burden, but it does not disclose other behaviors like rate limits, authentication needs, or error handling.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness5/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is extremely concise with two sentences, front-loading the core purpose and adding essential usage guidance and a behavioral trait without any fluff.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given the tool's complexity (11 parameters, output schema exists, sibling tool present), the description is mostly complete. It explains the tool's purpose and numbering, but could mention that results include standard fields (title, URL) though the output schema likely covers that.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema description coverage is 100%, so the baseline is 3. The description does not add any parameter-level information beyond what is already in the schema; it merely states the overall purpose.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states 'Fetch web results for a query using the Kagi Search API', which specifies the verb and resource. It further distinguishes usage for general search and when the user explicitly says 'fetch', helping differentiate from sibling kagi_extract (which likely extracts content from a URL).

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides clear usage context: 'Use for general search and when the user explicitly tells you to fetch results/information.' It implies when to use it but does not explicitly exclude alternative tools like kagi_extract or list when not to use it.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

  1. 2 tool updatesv1.0.1
    • First observedkagi_extract
    • First observedkagi_search_fetch
Deploy Server
One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

TDQS

A4.1/5.0

Scored across 2 tools

Disambiguation5/5

Each tool has a clearly distinct purpose: one extracts page content, the other fetches search results. There is no overlap or ambiguity between them.

Naming Consistency5/5

Both tools follow a consistent 'kagi_' prefix combined with a verb_noun pattern ('extract' and 'search_fetch'), making naming predictable and clear.

Tool Count3/5

With only two tools, the set is minimal but still covers the core search and extract functionalities. However, it feels thin compared to typical MCP servers, which might offer more diverse operations.

Completeness4/5

The server provides the essential operations for its domain: searching and retrieving content. While additional tools like summarization could be useful, the current set is reasonably complete for basic tasks.

Maintenance

ActivityStale
ResponsivenessResponsive

Resources

Related MCP Connectors

Related MCP Servers

  • Perplexity MCP Server

    RossH121
    A
    license
    A
    quality
    D
    maintenance
    An MCP server that enables Claude to perform web searches using Perplexity's API with intelligent model selection based on query intent and support for domain and recency filtering.
    6
    4
    MIT

  • kagi-server

    mikeysrecipes
    A
    license
    C
    quality
    D
    maintenance
    MCP server integrating the Kagi Search API to perform web searches using the kagi_search tool.
    1
    MIT

Appeared in Searches